Confluence Algebras and Acyclicity of the Koszul Complex

Transcription

1 Confluence Algebras and Acyclicity of the Koszul Complex Cyrille Chenavier To cite this version: Cyrille Chenavier. Confluence Algebras and Acyclicity of the Koszul Complex. Algebras and Representation Theory, 2016, < /s >. <hal v2> HAL Id: hal Submitted on 25 Jan 2016 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Confluence algebras and acyclicity of the Koszul complex CYRILLE CHENAVIER Abstract. The N-Koszul algebras are N-homogeneous algebras satisfying a homological property. These algebras are characterised by their Koszul complex: an N-homogeneous algebra is N-Koszul if and only if its Koszul complex is acyclic. Methods based on computational approaches were used to prove N-Koszulness: an algebra admitting a side-confluent presentation is N-Koszul if and only if the extra-condition holds. However, in general, these methods do not provide an explicit contracting homotopy for the Koszul complex. In this article we present a way to construct such a contracting homotopy. The property of side-confluence enables us to define specific representations of confluence algebras. These representations provide a candidate for the contracting homotopy. When the extracondition holds, it turns out that this candidate works. We make explicit our construction on several examples. Contents 1 Introduction 1 2 Preliminaries The Koszul complex Side-confluent presentations Extra-confluent presentations Confluence algebras and reduction operators The contracting homotopy in small degree Reduction operators and confluence algebras Reduction operators and side-confluent presentations The left bound of a side-confluent presentation Reduction pairs associated with a presentation Construction Extra-confluent presentations and reduction relations Examples The symmetric algebra Monomial algebras satisfying the overlap property The enveloping algebra of the Heisenberg Lie algebra Introduction An overview on Koszulness and N-Koszulness 1

3 Koszul algebras. Koszul algebras were defined by Priddy in [Pri70] as quadratic algebras which satisfy a homological property. A quadratic algebra is a graded associative algebra over a field K which admits a quadratic presentation X R, that is, X is a set of generators and R is a set of quadratic relations. If A is a quadratic algebra, the field K admits a left and right A-module structure induced by the K-linear projection ε : A K which maps any generator of A to 0. A quadratic algebra A is said to be Koszul if the Tor groups Tor A n,m K, K n is the homological degree and m is graduation induced by the natural graduation over A vanish for m n. A property of Koszul algebras is that the ground field K admits a Koszul resolution. The name of this resolution is due to the fact that it is inspired by ideas of Koszul see [Kos50]. Let A be a quadratic algebra and let X R be a quadratic presentation of A. We denote by KX and R the vector space spanned by X and the sub-vector space of KX 2 spanned by R, respectively. The Koszul complex of a A is the complex of free left A-modules: n+1 A J n n A Jn 1 where, for every integer n such that n 2, we have: J n = n 2 i=0 KX i R KX n 2 i. 4 A 3 J3 A R 2 A KX 1 A ε K 0, The differentials of the Koszul complex are defined by the inclusions of R in A KX, of J 3 in A R and of J n in A J n 1 for every integer n such that n 4. Then, a quadratic algebra is Koszul if and only if its Koszul complex is acyclic, that is, if and only if the Koszul complex of A is a resolution of K. Another characterisation of Koszulness was given by Backelin in [BF85] see also Theorem 4.1 in [PP05, chapter 2]: a quadratic algebra is Koszul if and only if it is distributive that means that some lattices defined with X and R are distributive. Moreover, Koszul algebras have been studied through computational approaches based on a monomial order, that is, a well founded total order on the set of monomials. In [Ani86], Anick used Gröbner basis to construct a free resolution of K see also [Ufn95, Section 3.8]. This resolution enables us to conclude that an algebra which admits a quadratic Gröbner basis is Koszul. In [Ber98], Berger studied quadratic algebras with a side-confluent presentation 1. The latter is a transcription of the notion of quadratic Gröbner basis using some linear operators. More precisely, we can associate with any quadratic presentation X R of A a unique linear projector S of KX 2. This projector maps any element of KX 2 to a better one with respect to the monomial order. The presentation X R is said to be side-confluent if there exists an integer k such that: S Id KX, Id KX S k = Id KX S, S Id KX k, where t, s k denotes the product sts with k factors. The algebra A k presented by: s 1, s 2 s 1, s 2 k = s 2, s 1 k, s 2 i = s i, i = 1, 2, is naturally associated with a side-confluent presentation. This algebra is the confluence algebra of degree k. In [Ber98, Section 5], Berger used specific representations of these algebras to construct a contracting homotopy for the Koszul complex of a quadratic algebra admitting a side-confluent presentation. This construction enables us to conclude that such an algebra is Koszul. N-Koszul algebras. Let N be an integer such that N 2. An N-homogeneous algebra is a graded associative algebra over a field K which admits an N-homogeneous presentation X R, that is, X is a set of generators and R is a set of N-homogeneous relations. In [Ber01] the notion of Koszul algebra 1 This notion corresponds to the one of X-confluent algebra in [Ber98]. However, we prefer to use our terminology because the property of confluence depends on the presentation. 2

4 is extended to the notion of N-Koszul algebra. An N-homogeneous algebra A is said to be N -Koszul if the Tor groups Tor A n,m K, K vanish for m l N n, where l N is the function defined by: l N n = kn, if n = 2k, kn + 1, if n = 2k + 1. We remark that a 2-Koszul algebra is precisely a Koszul algebra. Thus, the notion of N-Koszul algebra generalises the one of Koszul algebra. In the same paper, Berger defined the Koszul complex of an N-homogeneous algebra. Let X R be an N-homogeneous presentation of A. The Koszul complex of A is the complex of left A-modules: n+1 A Jn N n A J N n 1 4 A J N 3 3 A R 2 A KX 1 A ε K 0, where the vector spaces J N n are defined by: J N n = l N n N i=0 KX i R KX l N n N i. As in the quadratic case, this complex characterises the property of N-Koszulness: an N-homogeneous algebra is N-Koszul if and only if its Koszul complex is acyclic see [Ber01, Proposition 2.12]. This complex also find applications in the study of higher Koszul duality see [DV13]. Berger studied the property of N-Koszulness using monomial orders. As in the quadratic case, there exists a unique linear projector S of KX N associated with an N-homogeneous presentation of A which maps any element to a better one with respect to the monomial order. Then, a presentation is side-confluent if for every integer m such that N + 1 m 2N 1, there exists an integer k which satisfies: S Id KX m N, Id KX m N S k = Id KX m N S, S Id KX m N k. Contrary to the quadratic case, an algebra admitting a side-confluent presentation is not necessarily N- Koszul. Indeed, when the set X is finite, such an algebra is N-Koszul if and only if the extra-condition holds see [Ber01, Proposition 3.4]. The extra-condition is stated as follows: ec : KX m R R KX m KX m 1 R KX, for every 2 m N 1. We group these hypothesis in the following definition: Definition. Let A be an N-homogeneous algebra. A side-confluent presentation X R such that X is finite and the extra-condition holds is said to be extra-confluent. Our problematic. We deduce of the works from [Ber01] that the Koszul complex of an algebra A admitting an extra-confluent presentation is acyclic. However, there does not exist an explicit contracting homotopy for the Koszul complex of A. The purpose of this paper is to construct such a contracting homotopy. For the quadratic case, our contracting homotopy is the one constructed in [Ber98, Section 5]. Our results We present the different steps of our construction. Recall that an extra-confluent presentation needs a monomial order. Thus, in what follows, we fix a monomial order. For every integer m, we denote by X m the set of words of length m. 3

5 Reduction pairs associated with a presentation. In the way to construct our contracting homotopy, we will associate with any N-homogeneous presentation X R such that X is finite, a family P n,m = F n,m 1, F n,m 2, where F n,m 1 and F n,m 2 are linear projectors of KX m. The pair P n,m is called the reduction pair of bi-degree n, m associated with X R. We point the fact that the finiteness condition over X will be necessary to define the operators F n,m i. Moreover, these operators satisfy the following condition: for any w X m, F n,m i w is either equal to w or is a sum of monomials which are strictly smaller than w with respect to the monomial order. The linear projectors of KX m satisfying the previous condition are called reduction operators relatively to X m. The set of reduction operators relatively to X m admits a lattice structure we will recall it in Section 3.2. This structure plays an essential role in our constructions. A pair T 1, T 2 of reduction operators relatively to X m is said to be confluent if there exists an integer k such that we have the following equality in End KX m : T 1, T 2 k = T 2, T 1 k. Then, our first result is: Theorem. Let A be an N-homogeneous algebra admitting a side-confluent presentation X R, where X is a finite set. The reduction pairs associated with X R are confluent. The left bound of a side-confluent presentation. The reduction pairs associated with a sideconfluent presentation X R enable us to define a family of representations of confluence algebras in the following way: ϕ Pn,m : s 1, s 2 s 1, s 2 kn,m = s 2, s 1 kn,m, s 2 i = s i, i = 1, 2 End KX m, where the integer k n,m satisfies: F n,m 1, F n,m 2 kn,m = F n,m 2, F n,m 1 kn,m. s i F n,m i For every integers n and m we will consider a specific element in A kn,m : γ 1 = 1 s 2 s 1 + s 1 s 2 s s 2, s 1 2i+1, where the integer i depends on k n,m. The shape of this element will be motivated in Section 3.1. In Section 4.2 we will use the elements ϕ Pn,m γ 1 to construct a family of K-linear maps h 0 : A A KX, h 1 : A KX A R, h 2 : A R A J N 3, h n : A J N n A J N n+1, for n 3, where A is the N-homogeneous algebra presented by X R. The family h n n is called the left bound of X R. In Proposition 4.2.5, we will show that the left bound of X R is a contracting homotopy for the Koszul complex of A if and only if X R satisfies some identities. These identities are called the reduction relations. Extra-confluent presentations and reduction relations. Finally, we will show that the extracondition implies that the reduction relations hold. Then, our main result is stated as follows: 4

6 4.3.5 Theorem. Let A be an N-homogeneous algebra admitting an extra-confluent presentation X R. The left bound of X R is a contracting homotopy for the Koszul complex of A. Organisation In Section 2, we recall how we can construct the Koszul complex of an N-homogeneous algebra. We also recall the definition of an extra-confluent presentation. In Section 3.1, we make explicit our construction in small homological degree. In Section 3.2, we recall the definitions of confluence algebras and reduction operators. We also recall the link between reduction operators and representations of confluence algebras. In Section 4, we construct the contracting homotopy in terms of confluence. As an illustration of our construction, we provide in Section 5 three examples: the symmetric algebra, monomial algebras which satisfy the overlap properties and the enveloping algebra of the Heisenberg Lie algebra. Acknowledgement. The author wish to thank Roland Berger for helpful discussions. This work is supported by the Sorbonne-Paris-Cité IDEX grant Focal and the ANR grant ANR-13-BS CATHRE. 2 Preliminaries 2.1 The Koszul complex Conventions and notations. We denote by K a field. We say vector space and algebra instead of K-vector space and K-algebra, respectively. We consider only associative algebras. Given a set X, we denote by X and KX the free monoid and the vector space spanned by X, respectively. For every integer m, we denote by X m the subset of X of words of length m. We write V = KX. We identify KX m and the free algebra K X spanned by X to V m and to the tensor algebra T V over V, respectively. Let A be an algebra. A presentation of A is a pair X R, where X is a set and R is a subset of K X such that A is isomorphic to the quotient of K X by the two-sided ideal spanned by R. The latter is denoted by IR, and the isomorphism from A to K X /IR is denoted by ψ X R. For every f K X, we denote by f the image of f through the natural projection of K X over A. Let N be an integer such that N 2. An N-homogeneous presentation of A is a presentation X R of A such that R is included in V N. In this case, the two-sided ideal IR is the direct sum of vector spaces IR m defined by IR m = 0 if m < N, and IR m = m N i=0 V i R V m N i if m N, where R denotes the sub-vector space of V N spanned by R. An N-homogeneous algebra is a graded algebra A = m N A m which admits an N-homogeneous presentation X R such that for every integer m, ψ X R induces a K-linear isomorphism from A m to V m /IR m : A = m N A m K V V N 1 V N R V N+1 V R + R V We denote by ε : A K the projection which maps 1 A to 1 K and A m to 0 for every integer m such that m 1. 5

7 The construction of the Koszul complex. Let A be an N-homogeneous algebra and let X R be an N-homogeneous presentation of A. We write V = KX. We consider the family of vector spaces J N n n defined by J N 0 = K, J N 1 = V, J N 2 = R and for every integer n such that n 3 J N n = l N n N i=0 V i R V l N n N i, where the function l N : N N is defined by l N n = kn, if n = 2k, kn + 1, if n = 2k + 1. When there is no ambiguity, we write J n instead of J N n. Let n be an integer. For every w X l N n+1, let w 1 X l N n+1 l N n and w 2 X l N n such that w = w 1 w 2. Let us consider the A-linear map F n+1 : A V l N n+1 A V l N n. 1 A w w 1 w 2 Recall from [Ber01, Section 3] that the Koszul complex of A is the complex K, n+1 A J n n A Jn 1 2 A 1 J1 A ε K 0, where n is the restriction of F n to A J n. In particular, the map 1 is defined by 1 1 A x = x for every x X Remark. The two following remarks show that the Koszul complex is well-defined: 1. Let n be an integer. The vector space J n+1 is included in V l N n+1 l N n J n. Thus, the vector space F n+1 A J n+1 is included in A J n. 2. Let n be an integer such that n 1. The vector space J n+1 is included in R J n 1. Thus, the restriction of F n F n+1 to A J n+1 vanishes Example. We consider the enveloping algebra of the Heisenberg Lie algebra introduced in [AS87]. This is the 3-homogeneous algebra presented by x 1, x 2 x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2, x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2. This algebra is the minimal with respect to the number of generators example of Yang-Mills algebra introduced in [CDV02]. The map 2 : A R A V is defined by 2 1 A x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2 = x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2, and 2 1 A x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2 = x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2. The vector space J 3 = V R R V is the one-dimensional vector space spanned by v = x 2 x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2 + x 1 x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2 = x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2 x 1 + x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2 x 2. The map 3 : A J 3 A R is defined by 3 1 A v = x 2 x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2 + x 1 x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2. 6

8 2.2 Side-confluent presentations Through this section we fix an N-homogeneous algebra A and an N-homogeneous presentation X R of A. We assume that X is a totally ordered set. We write V = KX Reductions. For every integer m, the set X m is totally ordered for the lexicographic order induced by the order over X. For every f V m \ 0}, the leading monomial of f, denoted by lm f, is the greatest element of X m occurring in the decomposition of f. We denote by lcf the coefficient of lm f in the decomposition of f. Let } 1 R = lc f f, f R. Then, X R is an N-homogeneous presentation of A. Thus, we can assume that lc f is equal to 1 for every f R. For every w 1, w 2 X and every f R, let r w1fw 2 be the K-linear endomorphism of T V defined on the basis X in the following way: w1 lm f f w 2, if w = w 1 lm f w 2, r w1fw 2 w = w, otherwise. Taking the terminology of [Ber78], the morphisms r w1fw 2 are called the reductions of X R Normal forms. An element f T V is said to be a normal form for X R if rf = f for every reduction r of X R. Given an element f of T V, a normal form of f is a normal form g such that there exist reductions r 1,, r n satisfying g = r 1 r n f. In this case, we have f = g. The presentation X R is said to be reduced if, for every f R, lm f f is a normal form for X R and lm f is a normal form for X R \ f}. From this moment, all the presentations are assumed to be reduced Critical branching. A critical branching of X R is a 5-tuple w 1, w 2, w 3, f, g where f, g R and w 1, w 2, w 3 are non empty words such that: w 1 w 2 = lm f, and w 2 w 3 = lm g. The word w 1 w 2 w 3 is the source of this critical branching The operator of a presentation. Let S be the endomorphism of V N defined on the basis X N in the following way: lm f f, if there exists f R such that w = lm f, S w = w, otherwise. The operator S is the operator of X R. The presentation X R is reduced. Thus, S is well-defined and is a projector. The kernel of S is equal to R. If w X N is a normal form, then Sw is equal to w. If w is not a normal form, then Sw is a linear combination of words strictly smaller than w Definition. The presentation X R is said to be side-confluent if for every integer m such that 1 m N 1, there exists an integer k such that: Id V m S, S Id V m k = S Id V m, Id V m S k, where t, s k denotes the product sts with k factors. The Diamond Lemma [Ber78, Theorem 1.2] implies the following: 7

9 Proposition. Let A be an N-homogeneous algebra. Assume that A admits a side-confluent presentation X R. Then, the following hold: 1. Every element of T V admits a unique normal form for X R. 2. The set w, w X is a normal form} is a basis of A. 3. An element of T V belongs to IR if and only if its normal form is equal to 0. Proof. Let S be the operator of X R. Let w 1, w 2, w 3, f, g be a critical branching of X R. Let m be the length of w = w 1 w 2 w 3. The presentation X R being N-homogeneous, we have N + 1 m 2N 1. Thus, there exists an integer k such that: Id V m N S, S Id V m N k w = S Id V m N, Id V m N S k w. Hence, there exist two sequences of reductions r 1,, r n and r 1,, r l such that r 1 r n lm f f w 3 is equal to r 1 r l w 1 lm g g. We deduce from [Ber78, Theorem 1.2] that every element f T V admits a unique normal form for X R and that w, w X is a normal form} is a basis of A. Thus, the two first points hold. Let us show the third point. Let f be an element of T V and let f be its unique normal form. We write: f = i I λ i w i, where w i X are normal forms. Then, f is equal to i I λ iw i. From the second point, f is equal to 0 if and only if λ i is equal to 0 for every i I Lemma. Assume that the presentation X R is side-confluent. Let S be the operator of X R. For every integer m such that N + 1 m 2N 1, there exists an integer k such that: Id V m Id V m N S, Id V m S Id V m N k = Id V m S Id V m N, Id V m Id V m N S k. Moreover, for every w X m such that Id V m N Sw and S Id V m N w are different from w, we have: lm Id V m Id V m N S, Id V m S Id V m N k w = w. Proof. We write S 1 = Id V m N S and S 2 = S Id V m N. The presentation X R is side-confluent. Thus, there exists k N such that S 2, S 1 k is equal to S 1, S 2 k. The morphisms S 1 and S 2 being projectors, we show by induction that for every integer j we have: j 1 Id V m S 1, Id V m S 2 j = Id V m + 1 i S 1, S 2 i + S 2, S 1 i + 1 j S 1, S 2 j, i=1 j 1 Id V m S 2, Id V m S 1 j = Id V m + 1 i S 1, S 2 i + S 2, S 1 i + 1 j S 2, S 1 j. In particular we have: i=1 Id V m S 2, Id V m S 1 k = Id V m S 1, Id V m S 2 k. 8

10 Moreover, if w X m is such that S 1 w and S 2 w are different from w, then S 1 w and S 2 w are strictly smaller than w. We deduce from the relation k 1 Id V m S 1, Id V m S 2 k w = w + 1 i S 1, S 2 i + S 2, S 1 i w + 1 k S 1, S 2 k w, i=1 that lm Id V m S 1, Id V m S 2 k w is equal to w Example. We consider the presentation from Example of the enveloping algebra of the Heisenberg Lie algebra with the order x 1 < x 2. It was proven in [KVdB15, Theorem 6.3.2] that this presentation is side-confluent in fact, B.Kriegk and M.Van den Bergh have proven that any Yang-Mills algebra admits a side-confluent presentation. We propose there an other proof of this result. The operator S End V 3 of this presentation is defined on the basis X 3 by 2x 1 x 2 x 1 x 1 x 1 x 2, if w = x 2 x 1 x 1, Sw = 2x 2 x 1 x 2 x 1 x 2 x 2, if w = x 2 x 2 x 1, w, otherwise. This presentation admits exactly one critical branching: We have: x 2, x 2 x 1, x 1, x 2 x 1 x 1 2x 1 x 2 x 1 + x 1 x 1 x 2, x 2 x 2 x 1 2x 2 x 1 x 2 + x 1 x 2 x 2. S Id V, Id V S 2 x 2 x 2 x 1 x 1 = Id V S, S Id V 2 x 2 x 2 x 1 x 1 = x 2 x 1 x 2 x 1 2x 1 x 2 x 1 x 2 + x 1 x 1 x 2 x 2. Moreover, for every w X 4 which is different from x 2 x 2 x 1 x 1, we check that S Id V, Id V S 2 w is equal to Id V S, S Id V 2 w. Thus, we have: S Id V, Id V S 2 = Id V S, S Id V 2. For every w X 5 we check that S Id V 2, Id V 2 S 2 w and Id V 2 S, S Id V 2 2 w are equal. Thus, we have: S Id V 2, Id V 2 S 2 = Id V 2 S, S Id V 2 2. We conclude that the presentation from Example with the order x 1 < x 2 is side-confluent. 2.3 Extra-confluent presentations The extra-condition. Let A be an N-homogeneous algebra. Assume that A admits a sideconfluent presentation X R where X is a totally ordered finite set. Recall from [Ber01, Section 3] that the Koszul complex of A is acyclic if and only if the extra-condition holds. The extra-condition is stated as follows: KX n R R KX n KX n 1 R KX, for every 2 n N Definition. Let A be an N-homogeneous algebra. A side-confluent presentation X R such that X is finite and the extra-condition holds is said to be extra-confluent Remark. If N = 2, the extra-condition is an empty condition. Thus, in this case, the notions of extra-confluent presentation and side-confluent presentation coincide. An extra-confluent presentation has the following interpretation in terms of critical branching: 9

11 Proposition. Let A be an N-homogeneous algebra. Assume that A admits an extra-confluent presentation X R. Let w = x 1 x m be the source of a critical branching of X R. The word x m N x m 1 is not a normal form for X R. Proof. The presentation X R is N-homogeneous. In particular, we have N + 1 m 2N 1. If m = N + 1, there is nothing to prove. Thus, we assume that m is greater than N + 2. Let S be the operator of X R. We write S 1 = S Id m N and S 2 = Id m N S. The presentation X R is side-confluent. Thus, from Lemma 2.2.7, there exists an integer k such that Id V m S 2, Id V m S 1 k = Id V m S 1, Id V m S 2 k. We denote by Λ this common morphism. By hypothesis, S 1 w and S 2 w are different from w. From Lemma 2.2.7, lm Λw is equal to w. The image of Λ is included in im Id KX m S 1 im Id KX m S 2 that is, ker S 1 ker S 2. The latter is equal to R KX m N KX m N R. The presentation X R satisfies the extracondition. Thus, the image of Λ is included in KX m N 1 R KX. In particular, there exist w 1,, w l X m N 1, f 1,, f l R, x 1,, x l X and λ 1, λ l K which satisfy Λw = l λ i w i f i x i. i=1 Thus, lm Λw = w is equal to w i lm f i x i for some 1 i l. We conclude that x n N x m 1 is equal to lm f i. In particular, it is not a normal form Remark. Let A be the algebra presented by x < y xyx. This presentation is side-confluent. There is only one critical branching: xy, x, yx, xyx, xyx. The source xyxyx of this critical has length 5. We deduce from Proposition that the extra-condition does not hold. Let us check that the Koszul complex of A is not acyclic: the vector space J 3 is reduced to 0} and the map 2 : A R A V is defined by 2 1 A xyx = xy x. In particular, xy xyx belongs to the kernel of 2. Thus, we have a strict inclusion im 3 ker Example. We consider the presentation from Example The vector space V 2 R R V 2 is reduced to 0}. Then, the extra-condition holds. We conclude that the presentation from Example is extra-confluent. 3 Confluence algebras and reduction operators 3.1 The contracting homotopy in small degree Through this section we fix an N-homogeneous algebra A. We assume that A admits an extra-confluent presentation X R. This presentation is also fixed. We write V = KX. The aim of this section is to make explicit our contracting homotopy in small homological degree. The formal construction will be done in Section 4. We have to construct a family of K-linear maps h 1 : K A, and h n : A J n A J n+1, for 0 n 2, satisfying the following relations: 1 h 0 + h 1 ε = Id A and n+1 h n + h n 1 n = Id A Jn, for 0 n 2. 10

12 By assumption, the set X is finite. However, we will see that for the constructions of h 1, h 0 and h 1 this hypothesis is not necessary. From Proposition 2.2.6, every element f of T V admits a unique normal form for X R. This normal form is denoted by f. For every w X, we define [w] A V as follows: 0, if w is the empty word, [w] = w x, where w X and x X are such that w = w x. The map [ ] : X A V is extended into a K-linear map from T V to A V. Let w X be a non empty word. For every a A, the action of A on [w] is given by a.[w] = [fw], where f T V is such that a = f. In small homological degree, the Koszul complex of A is A V R R V 3 A R 2 A V 1 ε A K 0, where 1 is defined by 1 1 A v = v for every v V, 2 is defined by 2 1 A f = [f] for every f R and 3 is defined by 3 1 A g = v f where vf is a decomposition of g V R R V in V R. By definition of 3, 3 1 A g does not depend on the decomposition of g in V R The constructions of h 1 and h 0. The maps h 1 : K A and h 0 : A A V are defined by h 1 1 K = 1 A and h 0 a = [ f], where f T V is such that f = a. We have h 0 1 A = 0 and h 1 ε 1 A = 1 A. If A belongs to A m for m 1, we have ε a = 0 and 1 h 0 a = f. It follows that 1 h 0 + h 1 ε is equal to Id A The construction of h 1. Recall from Proposition that the algebra A admits as a basis the set w, w X is a normal form}. Thus, in order to define h 1 : A V A R, it is sufficient to define h 1 w x for every normal form w X and every x X. Moreover, h 1 has to satisfy the relation 2 h 1 w x = w x h 0 wx, 1 for every normal form w X and every x X. We define h 1 w x by Noetherian induction on wx. Assume that wx is a normal form. Then, let h 1 w x = 0. We have: h 0 wx = [ŵx] = [wx] = w x. Thus, Relation 1 holds. Assume that wx is not a normal form and that h 1 w x is defined and satisfies 1 for every normal form w X and every x X such that w x < wx. The word wx can be written as a product w 1 w 2, where w 2 X N is not a normal form. The presentation X R is reduced. Thus, there exists a unique f R such that f = w 2 ŵ 2. Let We have: h 1 w x = w 1 f + h 1 [w 1 ŵ 2 ]. 2 h 1 w x = [w 1 f] + 2 h 1 [w 1 ŵ 2 ] = [w 1 w 2 ] [w 1 ŵ 2 ] + 2 h 1 [w 1 ŵ 2 ]. 11

13 By induction hypothesis, 2 h 1 [w 1 ŵ 2 ] is equal to [w 1 ŵ 2 ] [ ŵ 1 w 2 ]. Hence, we have: 2 h 1 w x = [w 1 w 2 ] [ ŵ 1 w 2 ] Thus, Relation 1 holds. = w x [ŵx] = w x h 0 wx Remark. We consider the K-linear morphisms F 1 : A V V N, w 1 w 2 x w 1 w 2 x, F 1 1 : A V N A V, w 1 w 2 x w 1 w 2 x, F 1 2 : A V A V N, w 1 w 2 x w 1 ŵ 2 x. The inductive definition of h 1 implies that h 1 w x is equal to F1 F2 1 w x + F1 F2 1 F 1 1 F2 1 w x + F 1 F2 1 F 1 1 F2 1 2 w x +, where F 1 F2 1 F 1 1 F2 1 2k w x vanishes for k sufficiently large. In order to define h 2 we need the following: Lemma. Let A be an N-homogeneous algebra. Assume that A admits an extra-confluent presentation X R. Let w 1 X, w 2 X N 1 and x 1, x 2 X such that: 1. w 1 x 1 and x 1 w 2 are normal forms for X R, 2. w 2 x 2 is not a normal form for X R. The word w 1 x 1 w 2 is a normal form for X R. Proof. Assume that w 1 x 1 w 2 is not a normal form. By hypothesis, w 1 x 1 and x 1 w 2 are normal forms. Thus, there exist a right divisor u of w 1 and a left divisor v of w 2 such that ux 1 v has length N and is not a normal form. In particular, ux 1 w 2 x 2 is the source of a critical branching. From Proposition 2.3.4, the word x 1 w 2 is not a normal form, which is a contradiction. Thus, Lemma holds The construction of h 2. Recall from Proposition that the algebra A admits as a basis the set w, w X is a normal form}. Thus, in order to define h 2 : A R A J 3 it is sufficient to define h 1 w f for every normal form w X and every f R. Moreover, h 2 has to satisfy the relation 3 h 2 w f = w f h 1 2 w f, 2 for every normal form w X and every f R. We write w = w 1 x 1, f = w ŵ and w = w 2 x 2. We define h 2 w f by Noetherian induction on x 1 w 2. Assume that x 1 w 2 is a normal form. Let h 2 w f = 0. We have: h 1 2 w f = h 1 [wf] = h 1 [ww ] h 1 [wŵ ] = h 1 ww 2 x h 1 [wŵ ]. 12

14 From Lemma 3.1.4, ww 2 is a normal form. Thus, by construction of h 1, we have: h 1 ww 2 x = w f + h 1 [wŵ ]. We conclude that h 1 2 w f is equal to w f. Hence, Relation 2 holds. Assume that h 2 u g is defined and that E 2 holds for every normal form u X and g R such that yv < x 1 w 2, where y X and v X N 1 are such that u = u y and lm g = vz for u X and z X. We consider the two morphisms S 1 = S Id V and S 2 = Id V S. The presentation X R is side-confluent. Thus, from Lemma 2.2.7, there exists an integer k such that: Id V N+1 S 2, Id V N+1 S 1 k = Id V N+1 S 1, Id V N+1 S 2 k. We denote by Λ this common morphism. The image of Λ is included in ker S 1 ker S 2. The latter is equal to R V V R. Recall that we have: k 1 Id V N+1 S 2, Id V N+1 S 1 k = Id V N i S 1, S 2 i + S 2, S 1 i + 1 k S 2, S 1 k. Thus, we have: k 1 Λ = Id V N+1 S 2 + Id V N+1 S 2 1 i g i S 1, S 2, i=1 where g i S 1, S 2 denotes the product S 1 S 2 S 1 with i factors. In particular, there exist f 1,, f l R, x 1,, x l X and λ 1,, λ l K such that x i w i < x 1 w 2 where lm f i = w i y i and Then, let Λ xw = xf + l λ i x i f i. i=1 h 2 w f = w 1 Λ xw i=1 l λ i h 2 w 1 x i f i. i=1 We will show in Section 4 that Relation 2 holds Remark. We consider the K-linear maps F 2 : A V N V N+1, w 1 x w 2 w 1 xw 2, F 2 1 : A V N+1 A V N, w 1 xw 2 w 1 x w 2, F 2 2 : A V N A V N+1, w 1 x w 2 w 1 xw 2 Λ xw 2. The inductive definition of h 2 implies that h 2 w f is equal to F2 F2 2 w f + F2 F2 2 F 1 2 F2 2 w f + F 2 F2 2 F 1 2 F2 2 2 w f +, where F 2 F2 2 F 1 2 F2 2 2k w f vanishes for k sufficiently large. 13

15 Example. The construction of our contracting homotopy for the Koszul complex of the enveloping algebra of the Heisenberg Lie algebra is done in Section Reduction operators and confluence algebras We fix a finite set Y, totally ordered by a relation <. For every v KY \ 0}, we denote by lmv the greatest element of Y occurring in the decomposition of v. We extend the order < to a partial order on KY in the following way: we have v < w if either v = 0 or if lmv < lmw. In this section we recall some results from [Ber98] about reduction operators and confluence algebras Reduction operators. A linear projector T of KY is called a reduction operator relatively to Y if for every y Y, we have either T y = y or T y < y. We denote by Red Y the set of reduction operators relatively to Y Lattice structure. The set Red Y admits a lattice structure. To define the order, recall from [Ber98, Lemma 2.2] that if U, T Red Y are such that ker U is included in ker T, then im T is included in im U. Thus, the relation defined by T U if keru kert is an order relation on Red Y. We denote by L KY the lattice of sub-vector spaces of KY : the order is the inclusion, the lower bound is the intersection and the upper bound is the sum. To define the upper bound and the lower bound on Red Y, recall from [Ber98, Theorem 2.3] that the map θ Y : Red Y L KY, T kert is a bijection. The lower bound T 1 T 2 and the upper bound T 1 T 2 of two elements T 1 and T 2 of Red Y are defined in the following way: T1 T 2 = θ 1 Y kert 1 + kert 2, T 1 T 2 = θ 1 Y kert 1 kert Remark. The lattice Red Y admits Id KY as maximum and 0 KY as minimum Confluent pairs of reduction operators. A pair P = T 1, T 2 of reduction operators relatively to Y is said to be confluent if there exists an integer k such that: T 1, T 2 k = T 2, T 1 k. We will see in Section 3.3 the link between this notion and the side-confluent presentations Confluence algebras. Let k be an integer. The confluence algebra of degree k is the algebra presented by s 1, s 2 s 2 i = s i, s 1, s 2 k = s 2, s 1 k, i = 1, 2. This algebra is denoted by A k. Let us consider the following elements of A k : σ = s 1, s 2 k = s 2, s 1 k, γ 1 = 1 s 2 i I γ 2 = 1 s 1 i I s 2, s 1 i, s 1, s 2 i, λ = 1 σ + γ 1 + γ 2, 14

16 where I is the set of odd integers between 1 and k 1. We easily check that we have the following relations: γ i s i = γ i, for i = 1, 2, s i γ i = s i σ, for i = 1, P -representations of confluence algebras. Let P = T 1, T 2 be a confluent pair of reduction operators relatively to Y. Let k be an integer such that T 1, T 2 k = T 2, T 1 k. We consider the morphism of algebras ϕ P : A k End KY. s i T i 3a 3b The morphism ϕ P is called the P -representation of A k. Recall from [Ber98] that: ϕ P σ = T 1 T 2, ϕ P 1 λ = T 1 T 2. 4a 4b The left bound and the right bound. Let P = T 1, T 2 be a confluent pair of reduction operators relatively to Y. By definition of λ and from 3.2.6, we have: T 1 T 2 = T 1 T 2 + ϕ P γ 1 + ϕ P γ 2. 5 The morphisms ϕ P γ 1 and ϕ P γ 2 are called the left bound of P and the right bound of P, respectively. We end this section with the following: Lemma. Let P = T 1, T 2 be a confluent pair of reduction operators relatively to Y. Let W be a sub-vector space of KY. If W is included in ker T i for i = 1 or 2, we have: ϕ P γ i W = T 1 T 2 W. Proof. By definition, σ and γ i factorize on the right by s i. Hence, the restrictions of ϕ P σ and ϕ P γ i to W vanish. Thus, Lemma is a consequence of Relation Reduction operators and side-confluent presentations Let A be an N-homogeneous algebra. We suppose that A admits a side-confluent presentation X R where X is a totally ordered finite set. For every integer m, the set X m is finite and totally ordered for the lexicographic order induced by the order over X. We write V = KX Normal forms and the Koszul complex. In Lemma we will link together the Koszul complex of A and the reduction operators. In this way, recall from Proposition that every element f T V admits a unique normal form for X R, denoted by f. Let φ: T V T V. f f Recall from Proposition that for every f T V, we have f IR if and only if f = 0. Hence, φ induces a K-linear isomorphism φ from A to im φ. In particular, for every integer n, the morphism φ n = φ Id V l N n is a K-linear isomorphism from A J n to im φ J n. Thus, the Koszul complex K, of A is isomorphic to the complex of vector spaces K, n+1 im φ J n where n is equal to φ n 1 n φ 1 n. n im φ J n 1 2 im φ J 1 1 im φ ε K 0, 15

17 Definition. The complex K, is the normalised Koszul complex of A Lemma. 1. For every integer m, the restriction of φ to V m is a reduction operator relatively to X m and its kernel is equal to IR m. 2. Let n be an integer such that n 1. The morphism n is the restriction to im φ J n of the morphism ϕ n : m l N n V m T V defined by ϕ n V m = φ V m l N n 1 Id V l N n 1. Proof. Let us show the first point. The presentation X R is N-homogeneous. Thus, for every w X m, φw belongs to V m. In particular, the restriction of φ to V m is an endomorphism of V m. Let w X m. If w is a normal form, then φw is equal to w. In particular, φ V m is a projector. If w is not a normal form, then φw = ŵ is strictly smaller than w. Thus, φ V m is a reduction operator relatively to X m. Moreover, f is equal to 0 if and only if f belongs to IR. Thus, the kernel of φ V m is equal to IR m. Let us show the second point. Recall from that the differential n : A J n A J n 1 of the Koszul complex of A is the restriction to A J n of the A-linear map defined by: A V l N n A V l N n 1, 1 A w w 1 w 2 where w 1 X l N n l N n 1 and w 2 X l N n 1 are such that w = w 1 w 2. Thus, the map n is the restriction of the morphism which maps a word w of length m l N n to ŵ 1 w 2, where w 1 X m l N n 1 and w 2 X l N n 1 are such that w = w 1 w 2. The latter is equal to φ V m l N n 1 Id V l N n Lattice properties. Let S End V N be the operator of X R : S w = lm f f, if there exists f R such that w = lm f, w, otherwise. The properties of S described in imply that S is equal to θ 1 X N R. For every integers m and i such that m N and 0 i m N, we consider the following reduction operator relatively to X m : S m i = Id V i S Id V m N i. The kernel of S m i is equal to V i R V m N i. The presentation X R is side-confluent. Hence, the pair S 2N 1 i, S 2N 1 is confluent for every integers i and j such that 0 i, j N 1. We deduce from [Ber01, Section 3] and [Ber98, Theorem 2.12] that for every integer m such that m N, the sub-lattice of Red X m spanned by S m 0,, S m m N is confluent that is, the elements of this lattice are pairwise confluent and distributive that is, for every S, T, U belonging to this lattice, we have S T U = S U T U. 4 The left bound of a side-confluent presentation Through this section we fix an N-homogeneous algebra A. We assume that A admits an N-homogeneous presentation X R where X is a totally ordered finite set. This presentation is also fixed. We write V = KX. We consider the notations of j 16

18 4.1 Reduction pairs associated with a presentation For every integers n and m such that m l N n, we consider the following reduction operators relatively to X m : F n,m 1 = θ IR 1 X m m ln n V l N n, Id V m, if m < l N n + 1, F n,m 2 = θ 1 V m l N n+1 J X m n+1, otherwise. The pair F n,m 1, F n,m 2 is denoted by P n,m Definition. The pair P n,m is the reduction pair of bi-degree n,m associated with X R Lemma. Let n and m be two integers such that n 1 and l N n m < l N n + 1. Then, m l N n 1 is smaller than N 1 and F n 1,m 1 is equal to Id V m. Proof. First, we show that m l N n 1 is smaller than N 1. Assume that m is a multiple of N: m = kn. In this case, the hypothesis l N n m < l N n + 1 implies that n is equal to 2k. Thus, l N n 1 is equal to k 1N + 1. That implies that m l N n 1 is equal to N 1. Assume that m is not a multiple of N: m = kn +r with 1 r N 1. In this case, the hypothesis l N n m < l N n+1 implies that n is equal to 2k + 1. Thus, m l N n 1 = m kn is smaller than N 1. Let us show that F n 1,m 1 is equal to Id V m. The first part of the lemma implies that IR m ln n 1 is equal to 0}. Thus, the kernel of F n 1,m 1 is equal to 0}, that is, F n 1,m 1 is equal to Id V m Lemma. 1. Let n and let m be two integers such that m l N n + 2. We have: F n,m 1 = S m 0 S m m l N n Let n and m be two integers such that n 1 and m l N n + 1. We have: F n,m 2 = S m m l N n+1 Sm m N. Proof. By definition of, we have: m l N n+2 ker S m 0 S m m l N n+2 = = i=0 m l N n+2 = = i=0 m l N n+2 i=0 m l N n N i=0 ker S m i V i R V m N i = IE m ln n V l N n. V i R V m l N n N i V l N n V i R V m l N n N i V l N n 17

19 By definition of, we have: ker S m m l N n+1 Sm m N = = m N i=m l N n+1 m N i=m l N n+1 ker = V m l N n+1 S m i V i R V m N i = V m l N n+1 J n+1. l N n+1 N i=0 V i R V l N n+1 N i The map θ X m being a bijection, the two relations hold Theorem. Let A be an N-homogeneous algebra admitting a side-confluent presentation X R, where X is a finite set. The reduction pairs associated with X R are confluent. Proof. Let n and m be two integers such that m l N n. We have to show that the reduction pair of bi-degree n, m associated with X R is confluent. We proceed in four steps. Step 1. Assume that n = 0. We have P 0,0 = Id K, Id K. Thus, the pair P 0,0 is confluent. Let m be an integer such that m 1. The kernel of F 0,m 2 is equal to V m 1 J 1 = V m. Thus, F 0,m 2 is equal to 0 V m. In particular, the operators F 0,m 1 and F 0,m 2 commute, that is, they satisfy the relation 2 2. F 0,m 1, F 0,m 2 = F 0,m 2, F 0,m 1 Hence, the pair P0,m is confluent for every integer m. Step 2. Assume that n 1 and l n n m < l N n + 1. The pair P n,m is equal to F n,m 1, Id V m. Thus, the operators F n,m 1 and F n,m 2 commute. We conclude that the pairs P n,m such that n 1 and l n n m < l N n + 1 are confluent. Step 3. Assume that n 1 and l N n + 1 m < l N n + 2. From Lemma 4.1.2, the morphism F n,m 1 is equal to Id V m. In particular, the operators F n,m 1 and F n,m 2 commute. Thus, the pairs P n,m such that n 1 and l N n + 1 m < l N n + 2 are confluent. Step 4. Assume that n 1 and m l N n + 2. Lemma implies that F n,m 1 and F n,m 2 belong to the lattice generated by S m i, for 0 i m N. From the latter is confluent. Hence, the pairs P n,m such that n 1 and m l N n + 2 are confluent. 4.2 Construction Through this section, we assume that the presentation X R of A is side-confluent. From Proposition 2.2.6, every element f of T V admits a unique normal for X R. This normal form is denoted by f. We denote by φ the endomorphism of T V which maps an element to its unique normal form. We consider the notations of Section Lemma. For every integers n and m such that m l N n, the operator F n,m 1 is equal to φ V m l N n Id V l N n. Proof. From Point 1 of Lemma 3.3.3, the operator φ V m l N n Id V l N n is a reduction operator relatively to X m and its kernel is equal to IR m lnn V lnn. The map θ X m being a bijection, Lemma holds. 18

20 Lemma. Let n be an integer. Let h n : m l N n V m T V be the K-linear map defined by h n V = m ϕpn,m γ 1, where ϕ Pn,m γ 1 is the left bound of P n,m. The image of h n is included in im φ J n+1. Proof. Let m be an integer such that m l N n. endomorphism H of V m such that By definition of the left bound, there exists an ϕ Pn,m γ 1 = Id V m F n,m 2 F n,m 1 H. The image of F n,m 1 = φ V m l N n V l N n is equal to the vector space spanned by the elements with shape w 1 w 2 where w 1 X m l N n is a normal form and w 2 X l N n. Let G = θ 1 X l N n+1 J n+1. We have F n,m 2 = Id V m l N n+1 G. The latter implies that Id V m F n,m 2 = Id V m l N n+1 Id V l N n+1 G. We conclude that the image of ϕ Pn,m γ 1 is included in the vector space spanned by elements with shape wf where w X m l N n+1 is a normal form and f J n+1. This vector space is equal to im φ V m l N n+1 J n Definition. For every integer n, let h n = φ 1 n+1 h n φ n : A J n A J n+1, where φ n is the K-linear isomorphism between A J n and im φ J n defined in The family h n n is the left bound of X R Reduction relations. Let n and m be two integers such that m l N n. Then, we denote by K n m = im φ V m l N n J n. In particular, we have: im φ J n = m l N n K m n. We say that the presentation X R satisfy the reduction relations if for every integers n and m such that m l n n, the following equality holds: r n,m F n,m 1 F n,m 2 = F n 1,m K n m 2 K n m Proposition. Let A be an N-homogeneous algebra. Assume that A admits a side-confluent presentation X R where X is a finite set. The left bound of X R is a contracting homotopy for the Koszul complex of A if and only if X R satisfies the reduction relations. Proof. The left bound of X R is a contracting homotopy for the Koszul complex of A if and only if the family h n : im φ J n im φ J n+1 defined in Lemma is a contracting homotopy for the normalised Koszul complex of A. n. 19

21 From Point 2 of Lemma 3.3.3, the restriction of F n 1,m 1 = φ V m l N n 1 Id V lnn 1 to K n m is equal to the restriction of n to K n m. Thus, the family h n n is a contracting homotopy for K, if and only if for every n and m such that n 1 and m l N n, the following relation holds: ϕ P n,m s 1 ϕ Pn,m γ 1 + ϕ Pn 1,m γ 1 ϕ Pn 1,m s 1 K m n = Id m K. n From Relation 3b see page 15 and Relation 4a see page 15, we have: ϕ Pn,m s 1 ϕ Pn,m γ 1 = F n,m 1 ϕ Pn,m σ = F n,m 1 F n,m 1 F n,m 2. The image of F n,m 1 = φ V m l N n Id V l N n is equal to im φ V m l N n V l N n. Thus, K n m is included in im F n,m 1. In particular, the restriction of F n,m 1 to K n m is the identity map. We deduce that the left bound family of X R is a contracting homotopy for the Koszul complex of A if and only if the following relation holds: ϕ P n 1,m γ 1 ϕ Pn 1,m s 1 K m n = F n,m 1 F n,m 2 K n m From Relation 3a see page 15, ϕ Pn 1,m γ 1 ϕ Pn 1,m s 1 is equal to ϕ Pn 1,m γ 1. Thus, it is sufficient to show:. ϕ Pn 1,m γ 1 K m n By construction, K m n = F n 1,m is included in ker 2 K n m F n 1,m Hence, Relation 6 is a consequence of Lemma The following lemma will be used in the proof of Theorem 4.3.5: Lemma. Let n and m be two integers such that n 1 and l N n m < l N n + 1. The operators F n,m 1 and F n 1,m 2 commute. Proof. The pair P n,m being confluent, we deduce from Relation 4b see page 15 that F n 1,m 2 is polynomial in F n 1,m 1 and F n 1,m 2. Hence, it is sufficient to show that F n,m 1 commutes with F n 1,m 1 and F n 1,m 2. Let G = θ 1 X l N n J n. We have F n 1,m 2 = Id V m l N n G. Thus, F n,m 1 = φ V m l N n Id V l N n commutes with F n,m 2. Moreover, the morphism F n,m 1 respectively F n 1,m 1 maps a word w of length m to ŵ 1 w 2 respectively ŵ 1 w 2, where w 1 X m l N n and w 2 X l N n respectively w 1 X m l N n 1 and w 2 X l N n 1 are such that w = w 1 w 2 respectively w = w 1w 2. Thus, the two compositions F n,m 1 F n 1,m 1 and F n 1,m 1 F n,m 1 are equal to F n 1,m Extra-confluent presentations and reduction relations Through this section we assume that the presentation X R is extra-confluent. Our aim is to show that X R satisfies the reduction relations. In this way, we will show in Proposition that the extra-condition enables us to link together the reduction pairs associated with X R. We consider the notations of Section

22 Lemma. Let m, r and k be three integers such that m N +2, 2 k N 1 and r+k m N. Then, we have: 1. S m r 2. S m r+k = Sm r S r m S m r+k 1 S m r+k, S m r+k = Sm r+k 1 Sm r+k. Proof. Let us prove the point 1. The extra-condition implies the following inclusion: V r+k R V m N r k V r R V m N r V r+k 1 R V m N r k+1. Applying the bijection θ 1 X m, we have: S m r+k 1 Sm r S m r+k. By definition of the upper bound, we deduce that S r m induction on k, we obtain the first relation. S m r+k 1 Sm r+k is equal to Sm r S m r+k. By Let us prove the point 2. Recall from that the lattice spanned by S m 0,, S m m N is distributive. Thus, the left hand side of 2 is equal to S r m S m r+k S m r+k 1 Sm r+k. By the first point, for every integer i such that 0 i n 2, S m r+i Sm r+k is equal to Sm r+i Sm r+k, so it is greater than S m. By definition of the lower bound, the second relation holds. r+k 1 Sm k+r Lemma. Let n and m be two integers such that n 2 and l N n + 1 m < l N n + 2. We have: S m 0 S m m l N n+1 S m m l N n = Sm m l N n+1 Sm m l N n. 7 Proof. From Lemma 4.1.2, the hypothesis l N n + 1 m < l N n + 2 implies that m l N n is smaller than N 1. Assume that m is a multiple of N. The hypothesis l N n + 1 m < l N n + 2 implies that m is equal to l N n + 1. Thus, the left hand side of 7 is equal to S m 0 S m m l N n and the right hand side of 7 is equal to S m 0 S m m l N n. Hence, Relation 7 is a consequence of Lemma point 1. Assume that m is not a multiple of N. The hypothesis l N n + 1 m < l N n + 2 implies that n is even. Hence, the left hand side of 7 is equal to S m 0 S m m l N n 1 S m m l N n and the right hand side of 7 is equal to S m m l N n 1 Sm m l nn. If n is equal to 2 and m is equal to N + 1, these two morphisms are equal to S N+1 0 S N+1 1. If the couple n, m is different from 2, N + 1, Relation 7 is a consequence of Lemma point Lemma. Let n and m be two integers such that n 2 and m l N n + 2. Letting we have: T n,m = S m m l N n+2+1 Sm m l N n+1, T n,m F n 1,m 2 = F n,m 2. Proof. From Lemma 4.1.3, we have F n 1,m 2 = S m m l N n Sm m N, and F n,m 2 = S m m l N n+1 Sm m N. 21

23 The law being associative, it is sufficient to show: T n,m S m m l N n = Sm m l N n+1 Sm m l N n. 8 Assume that n is odd. We have l N n + 2 = l N n Hence, the left hand side of 8 is equal to S m m l N n+1 Sm m l N n. Moreover, l Nn+1 l N n is equal to N 1. Thus, Relation 8 is a consequence of Lemma point 1. Assume that n is even. We have l N n + 1 = l N n + 1. Hence, the left hand side of 8 is equal to S m m l N n+2+1 Sm m l N n 1 S m m l N n and the right hand side of 8 is equal to S m m l nn 1 Sm m l N n. Moreover, l Nn l N n is equal to N 1. Thus, Relation 8 is a consequence of Lemma point Proposition. Let A be an N-homogeneous algebra. Assume that A admits a side-confluent presentation X R. Then, the presentation X R satisfies the extra-condition if and only if for every integers n and m such that n 1 and m l N n + 1, we have: F n,m 1 F n 1,m 2 = F n,m 1 F n,m 2. Proof. For every integers n and m such that n 1 and m l N n + 1, let L n,m = F n,m 1 F n 1,m 2, R n,m = F n,m 1 F n,m 2. Step 1. Assume that n = 1. Fist, we show that: L 1,m = F 0,m 1. 9 The kernel of F 0,m 2 is equal to V m 1 J 1 = V m, that is, F 0,m 2 is equal to 0 V m. In particular, F 0,m 1 F 0,m 2 is equal to F 0,m 1. Moreover, the kernel of F 1,m 1 is equal to IR m 1 V and the kernel of F 0,m 1 is equal to IR m. The inclusion IR m IR m 1 V implies that F 0,m 1 is smaller than F 1,m 1. Hence, Relation 9 holds. Assume that m = N. The kernel of F 1,N 1 is equal to IR N 1 V = 0}, that is, F 1,N 1 is equal to Id V N. In particular, R 1,N is equal to F 1,N 2. Moreover, we have: F 0,N 1 = θ X N 1 IR N = θ X N 1 R, and F 1,N 2 = θ X N 1 J 2 = θ X N 1 R. Thus L 1,N and R 1,N are equal. Assume that m N + 1. From Lemma 4.1.3, we have: F 0,m 1 = S m 0 S m m N, F 1,m 1 = S m 0 S m m N 1, F 1,m 2 = S m m N. Thus, R 1,m is equal to F 0,m 1. We conclude that Proposition holds for n = 1 and m N. 22